Keywords: Spatial Reasoning, 3D
Abstract: We present a probabilistic approach for Spatial Relation Detection for 3D robotic perception. We exploit the state-of-the-art open set vocabulary object detectors and rich 3D geometric features to localize objects and their spatial relations. We carry out experiments and ablation studies on both real Spatial Sense dataset and synthetic semantic abstraction dataset and demonstrate challenges on the open-set vocabulary setting and effectiveness of our approach on both synthetic and real data.
Submission Number: 17
Loading